Effective schools do exist: low income children's academic performance in chile

Page created by Mark Leonard
 
CONTINUE READING
m a y o

003
     effective schools do exist: low income
  children’s academic performance in chile

         francisco henríquez _ bernardo lara _
             alejandra mizala _ andrea repetto
Effective Schools do Exist: Low Income Children’s Academic Performance in
                                        Chile

                                     Francisco Henríquez
                 Universidad Católica of Valparaíso and Ministry of Education

                                     Bernardo Lara
            Center for Applied Economics, Department of Industrial Engineering,
                                    University of Chile

                                     Alejandra Mizala1
      Center for Applied Economics, Department of Industrial Engineering and Center for
                    Advanced Research in Education, University of Chile

                                       Andrea Repetto
                       School of Government, Universidad Adolfo Ibáñez

                                           December 2009

Acknowledgements: We thank Lily Ariztía and Bernardita Muñoz who provided valuable information about
the SIP schools, as well as comments and suggestions. We also thank Magdalena Álvarez and Javiera
Martínez for valuable assistance. We are grateful to the SIMCE office at Chile’s Ministry of Education for
providing us with the data. Funding from Fondecyt (#1070316) and PBCT-CONICYT Project CIE-05 is
gratefully acknowledged. The authors bear sole responsibility for the views expressed in the paper.

1
  Corresponding author: República 701, Santiago, Chile. E-mail address: amizala@dii.uchile.cl.
Phone: (56-2) 978-4072. Fax: (56-2) 978-4011.
Abstract

The aim of this paper is twofold. First, we show that despite students’
disadvantaged backgrounds and despite not having more financial resources than
similar schools, there are schools in Chile that serve low income students and that
obtain superior academic outcomes. Second, we present qualitative evidence to
identify school and classroom processes that might explain these good results.
Specifically, we analyze a network of Chilean private voucher schools called
Sociedad de Instrucción Primaria (SIP). In the econometric analysis we use a
number of propensity score based estimation methods, to find that SIP students’
achievement is not due to observables or selection on measured variables. We also
performed a number of interviews to SIP schools and other neighboring schools.
Our qualitative analysis suggests that having children’s learning as a central and
permanent goal, an aim that is shared and that drives the community’s efforts,
seems to best summarize what makes SIP schools special.

Keywords: Educational quality, effective schools, school effect, propensity score methods.

                                                                                             2
I.      Introduction

At least since the publication of the Coleman report in 1966, educational researchers and

policymakers have struggled in their efforts to understand whether schools matter and

whether there are policies that lead to increased educational achievement.2 A large body of

literature provides mixed results on the role of school level resources such as classroom

size and teachers’ experience on student performance.3

At the same time, educational systems display great variation in the distribution of student

achievement as measured by test scores, dropout rates and college attendance (OECD,

2009). The enormous observed diversity in results is certainly correlated with students’

family backgrounds and experiences. However, it cannot be fully accounted for by

variation in household characteristics.

Chile’s educational system is no exception: it is also characterized by a large heterogeneity

in results, differences that are not completely associated with the socio-economic level of

the students (Mizala and Romaguera, 2005). Despite the sweeping reforms that were

introduced into Chile’s educational system over almost three decades ago, as of today

schools with stagnant scores still co-exist with schools that show continuous progress.

More specifically, starting in the early 1980s, Chile’s government began to provide

subsidies to schools to finance students wishing to attend private schools.4 It also delegated

2
  See Coleman et al (1966).
3
  Literature reviews include Hanushek (1989, 1997) for developed countries and Fuller (1990), Fuller and
Clarke (1994) and Hanushek (1995) for developing countries.
4
  Along this paper we will refer to these schools as private voucher or private subsidized schools. Although
the subsidy is paid directly to schools on a per-student basis, the system simulates a voucher-type demand
subsidy.

                                                                                                               3
public education to local governments (municipalities), tying the budget of schools to

enrollment. Although the reform intended to generate competition between schools, thus

promoting more efficient and better quality education services, large differences in the

performance of students attending different schools are still observed. This, in turn, has

consequences for income distribution, as primary and secondary school quality is an

important predictor of the access to university education, good jobs and high wages (Card

and Krueger, 1996).

This great and persistent heterogeneity in school performance has important research and

policy implications. If the lesson were that all the schools suffer from poor performance,

one could conclude that responsibility for these results does not depend so much on schools

but, instead, on policy makers, parents or even children. In contrast, if some schools

progress while others are left behind, different questions arise. What characteristics

distinguish high performing schools from those with low levels of student achievement?

What are the determinants of high performance? What lessons can be drawn from the

schools that show significantly better scores and that serve populations with similar

characteristics? And can these lessons be applied to improve the performance in schools

with poor achievement?

This paper intends to add evidence on the strategies that are effective for students from

disadvantaged economic backgrounds. That is, we first provide statistical evidence that

there exist schools attending low income children that consistently produce outstanding

results. Then we provide insight on the reasons behind the relative success of these schools

on the basis of information gathered in a number of interviews.

                                                                                             4
Specifically, we concentrate the analysis on the relative performance of children attending

schools that belong to the Sociedad de Instrucción Primaria (SIP), a non-for profit

organization that serves low income students in Santiago, Chile, since 1856. The network

consists of 17 private voucher schools attending about 18 thousand students. SIP schools

are also known as Matte schools, in honor of the founding family. Students at these schools

stand out because of their systematic superior performance in national standardized SIMCE

tests (Educational Quality Measurement System). Despite their students’ disadvantaged

family background and despite not having more financial resources than similar schools,

SIP schools obtain math scores in the national standardized achievement tests that are 102

percent of one standard deviation higher than those obtained by public/municipal schools

and 69 percent of one standard deviation higher than private voucher schools in Santiago.

Similar results are found for language test scores (see Table 1). Furthermore, the

performance of SIP schools is very similar to that of private non-voucher schools that

typically serve the most elite families in Chile. In other words, SIP’s education seems a cost

effective way of improving school quality.

In this paper we intend to answer the types of questions above through an analysis of SIP

schools. That is, the main goal of this paper is to shed light on the factors that contribute to

a better education for low-income students attending Matte schools. The experience of

students at these schools shows that it is possible to create an environment in which low-

income students can achieve high academic standards. The lessons of this analysis can help

improve the performance in schools with poor achievement.

In the first part of this paper we analyze whether the superior performance of Matte

students can be explained by observable factors, such as schools, students and family

                                                                                                   5
characteristics. Using a number of propensity score based estimation methods we find that

after controlling for observables and selection on measured variables, SIP students perform

much better in SIMCE tests than students at private voucher and municipal schools. In fact,

we show that they even perform better than those attending private non voucher schools;

i.e, students enrolled at the schools that attend the children of the elite families in Chile.

The estimated effect is not only statistically significant, but economically relevant. That is,

we find differences in test scores up to 0.9 standard deviations compared to students in

public/municipal schools, and up to 0.7 standard deviations compared to students in private

voucher schools. Thus, compared to other education interventions, SIP appears as quite

effective.5

After showing that SIP students’ superior performance is not spurious, we look further into

understanding the unmeasured variables that explain differences in performance. In order to

do so, we conducted a series of interviews with SIP schools’ principals and with the

principals of schools that compete with them. For each Matte school and within the same

municipality, we chose schools that are close in terms of the average socioeconomic

background of the students, enrollment and the availability of financial resources. Our

interviews reveal a number of relevant differences across schools. Perhaps the most striking

difference is that SIP schools have had children’s learning as their primary and permanent

goal, an aim that is shared and that drives the community’s efforts.

5
  For instance, according to Angrist et al. (2002), the introduction of vouchers targeted to low income students
in Colombia increased academic achievement in 0.2 standard deviations.

                                                                                                              6
The remainder of the paper is organized as follows: Section II provides a general view of

the Chilean Educational System and reviews the related literature. Then Section III

introduces SIP schools and analyzes the robustness of its superior performance to

controlling for observables and for selection on measured variables. Section IV describes

the results of our interviews. Finally, Section V concludes.

II.     The Chilean educational system

In the early 1980s sweeping reforms were made to Chile’s educational system: the public

sector school system was decentralized and school management was delegated to local

government –municipal-- authorities.6 The reform also paved the way for the private sector

to enter the market as a provider of education by introducing a voucher-type demand

subsidy to finance municipal and private voucher schools. The voucher, which is paid

directly to schools on a per-student basis, is intended to cover running costs and generate

competition between schools to attract and retain students, thus promoting more efficient

and better quality education services.7 Since 1993, private voucher schools have been

allowed to charge tuition on top of the voucher that is received from the government.

Moreover, a system of standardized tests for measuring educational attainment, known as

the SIMCE, was established in order to evaluate the success of the reforms, inform parents

about the quality of their schools and provide a basis for future political decision.

6
 For a more detailed description of the Chilean educational system see Mizala and Romaguera (2000).
7
 The monthly voucher for primary school students amounted to slightly over $51 in 2002. As a reference, the
average monthly after tax wage in Chile was about $570 according to the 2003 CASEN Household Survey.
We refer to 2002 data because the empirical section of this paper uses data from the 2002 SIMCE test in the
Metropolitan Region of Santiago.

                                                                                                         7
As a result of these reforms, the number of new schools in the private sector has increased

   rapidly over the past twenty years. In 1985, there were 2,643 private voucher schools in

   Chile; this number grew to 3,640 in 2002 and to 5,054 in 2007. The resulting three-legged

   school system comprises of:8

1. Private non-voucher schools, which are financed by fees paid by parents and accounted for

   8.5 percent of all students in Chile and 12.7 percent of all students in the Metropolitan

   Region of Santiago in 2002.

2. Private voucher schools, which are financed by the per-student subsidy provided by the

   government and may be co-financed by monthly fees paid by parents. These schools are

   owned and ran by the private sector. These accounted for 37.8 percent of total enrollment in

   Chile and for 47.6 percent of students in Santiago in 2002.

3. Public schools, which are financed by the voucher but are owned and managed by

   municipal authorities. They represented 52.1 percent of the enrollment in Chile and 37.6

   percent of students in Santiago in 2002.

   The system has proven highly effective in terms of coverage. As a result of the voucher

   system the secondary school enrollment increased from 65% in the early 1980s to almost

   90% by 2003. Graduation rates have also increased sharply, whereas secondary school

   dropout rates have declined.9 Nevertheless there is still a substantial gap between the

   8
    The remaining of the school population attends schools run by educational corporations linked to business
   organizations or professional-technical secondary schools.
   9
     Primary school coverage had already reached levels over 95% before the reform (Mizala and Romaguera,
   2005).

                                                                                                            8
average test scores of Chilean students and those of students from other countries in

international assessments of student learning (Martin et al., 2004, and OECD, 2004).

A number of papers have examined the performance of the Chilean students based on

SIMCE data. Most of these papers have studied the relative effectiveness of private versus

public schools, while others have investigated the effect of school competition on student

academic outcomes. Using a number of statistical techniques --including OLS, Heckit,

propensity score based methods and change-in-changes--, most studies using individual-

level data, have found that students attending private voucher schools have higher

educational outcomes than those from public schools.10 However, papers that have tried to

identify the effect of inter-school competition on students’ achievement in Chile have

reached mixed results.11

More recently, a number of qualitative case studies have discussed the characteristics that

distinguish high performing schools in poor areas. These papers have found that variables

correlated with high performance are not easy to measure and usually are not observed by

an econometrician. Instead, they have found that effective schools have, among other

characteristics, clear goals and objectives, committed and highly professional teachers, an

instructional leader at the school who evaluates and supports teachers in their classroom

practices, a well-respected principal, clear rules, and multiple approaches to reach a

10
   Mizala and Romaguera (2001), Sapelli and Vial (2002 and 2005), Anand, Mizala and Repetto (2008) and
Lara, Mizala and Repetto (2009) find a statistically significant effect of private voucher education on SIMCE
test scores that ranges from 4% to 45% of one standard deviation. However, McEwan (2001) finds that non-
religious private voucher schools, --that is, the typical private voucher schools-- are less effective than
municipal schools. Similarly, when using changes- in-changes techniques, Lara et al (2009) find no
difference in the performance of municipal and private voucher schools students.
11
   Gallego (2002 and 2006), Hsieh and Urquiola (2006), Auguste and Valenzuela (2003) and Gallego and
Hernando (2008).

                                                                                                            9
heterogeneous student population (Belleï, Muñoz, Pérez and Raczynski, 2004; Raczynski

and Muñoz, 2005; Eyzaguirre and Fontaine, 2008, Pérez and Socias, 2008).

Two other papers have addressed SIP students’ performance. García and Paredes (2006)

use a case study approach to describe a series of practices performed by this network and

analyze the cost of implementing them. Their findings suggest that with an appropriate

management, low income students can perform well with the resources brought up mainly

by the voucher system, complemented by a smaller fraction provided by the fee paid by

parents.

In a related paper, Mackenna (2006) uses propensity score matching and Heckman

selection correction to find that SIP students outperform those attending public and other

private voucher schools. The differences are large ranging between 70 percent and 80

percent of one standard deviation when compared to municipal students, and between 56

percent and 72 percent of one standard deviation when compared to other private voucher

schools. Moreover, and contrary to the literature that compares private subsidized and

public schools, the paper does not find evidence of selection in SIP schools.

In the two stage estimation, Mackenna (2006) uses as instruments the number of SIP

schools and the number of municipal schools in the provincial department. That is, the

paper assumes that school selection is affected by local school supply but that these school

densities do not affect achievement. Although this is a standard identification strategy in the

literature, it is likely that private schools are not randomly assigned across neighborhoods.

That is, private schools locate in communities with families characterized by unmeasured

variables that may positively influence achievement.

                                                                                             10
Moreover, when using propensity score matching, it is important to account for the fact that

since there are only a number of SIP schools, the treatment group has a small number of

observations relative to the control group. This could entail lack of overlap on the

covariates, and thus bias and large variance in the estimated average treatment effect

(Crump, et al., 2009).

In this paper we perform our analysis using propensity score based methods to estimate the

effect of SIP education and to circumvent the problems described above. The estimation

methods are based upon two main assumptions: unconfoundedness and overlap. The first

assumption implies that participation in the treatment program does not depend on the

outcome after controlling for differences in observed variables, such as socioeconomic

status. It is a very controversial assumption (Imbens and Wooldridge, 2009), but still very

popular, especially since Dehejia and Wahba (2002) showed good results in comparing

experimental data and matching results in the evaluation of a training program.

The second assumption states that individuals should have positive probabilities of being

observed in both treatment and control groups. Although it seems a less controversial

assumption, recent literature has proposed new methods to improve overlap. In particular,

in order to overcome the lack of overlap due to the smaller number of individuals in the

treatment group vis a vis the control group, in this paper we implement the trimming

strategy proposed by Crump et al (2009).

We use three estimation methods based upon propensity scores. First, we use matching,

estimating the effect using as the counterfactual the observation with the closest propensity

score. A second method is propensity score weighting. This method weights the

                                                                                            11
observations using the propensity score and the treatment status. The idea is to balance the

sample between treated and non treated individuals based on the probability of treatment.

Specifically, we use the inverse probability weighting (IPW) estimator proposed by Hirano,

Imbens and Ridder (2003). Finally, we use the propensity score weighted regression,

introduced by Robins and Rotnizky (1995), Robins et al (1995), and Robins and Ritov

(1997), which allows us to directly account for the correlation between covariates and

outcomes.

III.   SIP schools and their relative performance

SIP schools are private voucher schools. Out of a total of 17 schools, 15 serve children at

the primary level. Our study is based on the performance of children attending these

primary level schools.

The direction of the schools is under central management, comprised of the offices of

General Management, Accounting, Human Resources, Computing Services, and the

Pedagogical Department. These central offices delineate the aims and mission, leaving each

school enough autonomy to attain the required goals in different ways.

The Pedagogical Department plays a key role in SIP schools. It defines the standards, goals

and the progress expected from each school. It also monitors the performance of each

school, and evaluates and measures the students’ achievement. Furthermore, it organizes

the remedial measures in case the goals are not met. A unit of the Pedagogical Department,

the Family Orientation Unit, is in charge of ensuring the participation of families in the

educational process.

                                                                                              12
SIP schools are mainly financed by the state voucher, which represents about 80% of all

revenue. About 20% of revenues come from fees charged to parents, and from donations,

typically targeted to specific projects, such as libraries and new infrastructure.

Table 1 shows the outstanding performance of SIP’s 4th graders in the 2002 Math and

Language SIMCE standardized tests compared to children in all types of schools –

public/municipal, private voucher and private non voucher schools. Students at SIP schools

not only outperform students attending public and private voucher schools; their

performance is quite similar to those at private non voucher schools which typically serve

only the most elite families in Chile. SIP students also show a lower level of heterogeneity

in results, as suggested by the lower standard deviation of SIMCE scores. Figure 1 displays

the distribution of SIMCE scores in the Metropolitan Region of Santiago, urban areas,

compared to the distribution for SIP students. Statistical tests show that SIP distributions

first-order stochastically-dominate the aggregate distribution of SIMCE scores.12 Figure 2

decomposes the aggregate distribution into the distributions by school type. The graphs

show that SIP students achieve results that better resemble the performance of students at

elite schools in Chile rather than the performance of students at public and at other private

voucher schools.

In what follows we show that the superior performance of students enrolled at SIP cannot

be fully attributed to observable variables and selection on measured variables. In later

12
   Our econometric results are based on the 2002 SIMCE test. However, SIP students have systematically
displayed superior performance in every SIMCE test.

                                                                                                   13
sections we look into qualitative differences that might explain the higher achievement of

SIP students.

a. Comparing the performance of children at SIP and at other schools in the

Metropolitan Region of Santiago: data and methodology

In this section we study whether the gap in performance between SIMCE and otherwise

similar students is robust to controlling for selection on observables and for the

characteristics of schools, students and their families. To answer this question, we use the

2002 SIMCE test scores, taken by all 4th grade children in Chile. We use the 4th grade test

since most SIP schools offer primary education only. We do not use the 8th grade SIMCE

test because many of the best students at these schools move to high achievement

secondary schools at the end of 6th grade.

The SIMCE data set is complemented by data from a questionnaire that is answered by

the parents of students that participated in the test. This questionnaire provides

information on the socio-economic characteristics of each student, such as their family

income and the education of the parents, as well as their educational history. In addition,

we use data from the Ministry of Education and the Under-secretary of Regional

Development to calculate the per pupil resources that were available to each school.

Modifications were made to some of the variables in the database in order to make them

compatible with our analyses. First, we only analyze students that reside in the urban areas

of the Metropolitan Region of Santiago because this is the region in Chile where SIP

schools are. Second, on the parental questionnaire, parents reported the highest level of

education that they had attended. These levels were converted into the corresponding

                                                                                               14
number of years they had been in formal education: the maximum time a parent could

spend in basic education is 8 years, high school is 12 years, technical institute is 14 years,

professional institute is 16 years, college is 17 years, a masters degree is 19 years, and a

doctoral degree is 23 years. Third, parents also reported their monthly income as a range

(for example, a parent could report that their income is between 400,000 to 500,000 pesos).

These ranges were replaced with the midpoint of the range, which means in the prior

example the parent would have an income of 450,000 pesos. Furthermore, the income was

divided by 1,000 to simplify the interpretation of results.

Table 2 provides a summary of the main characteristics of schools, students and their

families in our data base. Students at public schools belong to less advantaged

households. Their household income is lower, parents have reached lower levels of

education and have access to fewer educational resources at home – measured by the

number of books and the availability of computers at home—and financial resources at

school. Students attending SIP schools seem quite similar to those attending other private

voucher schools in a number of relevant characteristics such as household income and

maternal education. Nevertheless, they are more likely to have a computer at home and on

average own a larger number of books. This gap might capture differences in the

motivation of parents. In addition, teachers at SIP schools have more years of experience.

Finally, Table 2 shows how different students attending private non voucher schools are.

Families have incomes that are almost 5 and almost 8 times larger than families whose

children attend private voucher and municipal schools respectively. Mothers of children at

private non voucher schools have on average 4 to 6 more years of completed education,

and there are twice as many books at home. We do not have information on the total

                                                                                               15
amount of resources per child at school in these elite schools, but tuition charges can

reach up to 10 times the state voucher.

The main goal of this section is to evaluate the treatment effect of attending a SIP school on

students’ performance. The main challenge is to address selection bias – students that

attend SIP schools may have characteristics that are correlated with academic achievement-

- since parents are free to choose a school for their children, and at the same time, private

schools are free to choose the students they admit. To address this problem, we use a

number of propensity score based techniques identifying groups of similar students

attending different types of schools. In all cases we define as the treatment having attended

a SIP school, and then compare the outcome of these students in the treatment group with

those attending either municipal, private voucher or private non voucher schools. That is,

we perform three sets of comparisons across treatment and control groups, with each

control group defined by students attending a different type of school.

More specifically, as any paper that intends to identify treatment effects with a treated-

nontreated formulation, this study faces the classical problem of only observing the post-

treatment outcome in the observed treatment status. As in the usual Rubin model (Rubin,

1974), we only observe:

                                    Yi = Yi 1 ⋅ Di + Yi 0 ⋅ (1 − Di )

where

Yi = Observed post-treatment outcome for individual i.

Yi1 = Post-treatment outcome for individual i in case of receiving the treatment.

                                                                                                16
Yi0 = Post-treatment outcome for individual i in case of not receiving the treatment.

Di = Dummy indicating whether individual i did receive the treatment or not.

Propensity score matching is a technique used for non-experimental data to identify a

control group that exhibits the same distribution of covariates as the treatment group.

Propensity score matching is often used by statisticians and is becoming increasingly

popular among economists as a method to measure the impact of training programs. The

most common application of propensity score matching is to estimate the impact of job

training programs (Heckman et al., 1997; Dehejia and Wahba, 2002). This paper applies a

similar methodology to estimate the impact of private school education on academic

achievement.

There are three important assumptions that make propensity score matching a feasible

model. The first is the statistical independence of (Y0, Y1) and D conditional on X:

       Assumption 1: (Y0, Y1) ⊥ D | X

where Y0 is the SIMCE score of a student in the control group, Y1 is the SIMCE score of a

student in a SIP school, D is the type of school the student attends, and X is the students’

characteristics. This first assumption allows us to use the outcome of a student in the

control group with X characteristics as a proxy for E(Y0 | D= 1, X). However, matching

students based on X is a complex process given the high dimensionality of X. According to

Rosenbaum and Rubin (1983), an alternative to matching based on X is to use the

propensity score, which they define as "the conditional probability of assignment to a

particular treatment given a vector of covariates." Rosenbaum and Rubin argue that if the

                                                                                               17
matched observations have homogenous propensity scores, then they will also have the

same distribution of X. Let the propensity score be denoted by P(X) = Pr(D=1 | X). In

situations in which P(x) is not known, it can be estimated by models such as the probit or

logit.

The second assumption for propensity score matching is

         Assumption 2: 0 < P(x) < 1

If P(x) equals 1, then students with those characteristics always attend a SIP school and

therefore no match in the control group can be found. The same logic applies if P(x) equals

0. As a result, the assumption that P(x) lies between 0 and 1 is an important condition to

guarantee that matches can be found for all students. This condition is typically achieved by

imposing common support. In this paper we go further and follow Crump et al (2009) in

implementing a strategy for selecting the group with overlap that minimizes the asymptotic

variance of the efficient estimator of the average treatment effect. This strategy, called

trimming, is one way of solving the lack of overlap in the covariate distributions between

treatment and control groups due to limited number of observations. This problem can lead

to imprecise estimates that are sensitive to the specification chosen (Imbens and

Wooldridge, 2009). In practice, the strategy is a very simple one: it discards observations

that are less than α away from zero and one. As in Crump et al (2008), we use α equal to

0.1.

The practical implementation is as follows. First, we estimate the propensity score. Then

we trim the observations with estimated propensity score below 0.1 or above 0.9. With this

new sample, we reestimate the propensity score to then apply the estimation methods

                                                                                              18
described below. Figures 3a, 3b and 3c show pre- and post-trimming propensity score

distributions for the treatment group (SIP schools) and the different control groups. In

general, propensity scores are very close to zero and do not seem to have an adequate

common support in the pre-trimming histograms, a situation that considerably improves

post-trimming.13

The propensity score is estimated on the basis of a number of observable variables, but we

must also consider the unobservable characteristics that are typical of students in private

schools. The third assumption states that the unobserved characteristics that are captured by

the error term, U0, have the same distribution regardless of whether the student is in the

treatment or control group.

           Assumption 3: E(U0 | D = 1, P(X)) = E(U0 | D=0, P(X))

As emphasized by Heckman et al. (1997), this assumption does not imply that E(U0 | P(X))

= 0; rather, it assumes that the distribution of the unobservables is the same for the

treatment and control groups.

Assumptions 1 and 3 are important for the consistency of the estimator. If these

assumptions are not valid, then our results may be biased. For instance, if SIP schools

admit students on the basis of unobserved measures or if the most motivated parents –in

unmeasured ways-- enroll their children to SIP schools, then our estimated effects may

overestimate the true effect of SIP education.

13
     Tables with descriptive statistics after trimming are available from the authors upon request.

                                                                                                      19
Based on these assumptions, we implement three estimators of the effect of SIP education.

The first one is the one-to-one estimator with replacement. The second one is propensity

score weighting. This method weights the observations using the propensity score and the

treatment status. The idea is to balance the sample between treated and nontreated

individuals based on the probability of treatment (Imbens and Wooldridge, 2009).

Specifically, we use the inverse probability weighting (IPW) estimator proposed by Hirano,

Imbens and Ridder (2003):

                           N     Di ⋅ Yi      N         Di       N (1 − D ) ⋅ Y              N     1 − Di
                          ∑                   ∑                −∑           i      i
                                                                                            ∑
                          i =1   ρˆ ( X i )   i =1   ρˆ ( X i ) i =1 1 − ρˆ ( X i )         i =1 1 − ρ
                                                                                                     ˆ(X i )

where ρˆ ( X i ) stands for the estimated propensity score.

The third propensity score based method considered in this paper allows us to directly

account for the correlation between covariates and outcomes. The method is the propensity

score weighted regression, introduced by Robins and Rotnizky (1995), Robins et al (1995),

and Robins and Ritov (1997).14 Specifically, we estimate the propensity score and obtain

its predictor ρˆ ( X i ) . Separately we fit a regression of the test score on the baseline variables

for the treatment group, mˆ 1 ( X i ) , and the control group, mˆ 0 ( X i ) .Then we estimate the effect

of the treatment as

        1    N
                 Di ⋅ Yi − ( Di − ρˆ ( X i )) ⋅ mˆ 1 ( X i ) 1        N
                                                                             (1 − Di ) ⋅ Yi + ( Di − ρˆ ( X i ))mˆ 0 ( X i )
            ∑                                               −        ∑
        N   i =1                ρˆ ( Xi )                     N       i =1                  1 − ρˆ ( X i )

14
  Further details on this method can be found in Imbens (2004) and Imbens and Wooldridge (2009). A
generalization of this method is described in Wooldridge (2007).

                                                                                                                               20
This method also has a double-robustness feature: it provides consistent estimators if either

the probability of treatment or the outcome regressions are incorrectly specified, implying

some safeguard against model misspecification.

Our implementation of the inverse probability weighting (IPW) and double-robust (DR)

estimators follows the steps suggested by Emsley et al. (2008).15

b. The decision to attend a SIP school: results

The first stage of our strategy is the estimation of the propensity score using all

observations. Then, following Crump et al (2009), we discard observations with high or

low predicted propensity score to then reestimate the probability that a student attends a SIP

school. Table 3 provides these estimates before and after trimming the relevant samples.

When the alternative is attending a public school, our results suggest that family income,

maternal education, household size, age, having attended pre-school, the number of books

at home, the availability of a computer at home, the school financial resources and the

teachers per pupil ratio, all have a statistically significant effect on the probability of

attending a SIP school. However, when compared to other private voucher schools, family

income, household size, age, preschool education and teachers per pupil show no

statistically relevant effect. The number of years of teacher experience now does. Finally,

when comparing to students attending private non voucher schools, maternal education,

resources at home and at school and the child’s age are the variables that have a statistically

significant effect on the probability of attending a SIP school.

15
  The estimations are performed using the dr Stata command, which gives the sandwich estimators of the
errors.

                                                                                                         21
As expected, when the sample is rebalanced and the remaining individuals are more alike,

the model losses explanatory power. However, the balancing allows for the construction of

treated and non treated groups that gain in comparability, as suggested by the distributions

of propensity scores shown in Figures 3a, 3b and 3c.

c. Student performance by school type

We next estimate the second stage of our model in order to estimate the effect of attending

a SIP school. Table 4 summarizes our results. The first panel compares the SIMCE scores

of students enrolled at a Matte school relative to similar students enrolled at a public

school. The differences are very large and always statistically significant. They range from

0.48 to 0.57 standard deviations in the language test, and from 0.74 to 0.87 standard

deviations in the math test depending on the estimation method.

When compared to similar students at other private voucher schools, the effect of SIP

education is estimated to be somewhat smaller but still very large. The smallest gains for

the language test are predicted by the double-robust model, which estimates an effect of

0.25 standard deviations; the largest is 0.44 standard deviations predicted by the matching

approach. Similarly, in the math test the smallest gains are also predicted by the double-

robust model (0.45 standard deviations) and the largest by the matching method (0.70

standard deviations).

Finally, we estimate the effect of attending a SIP school relative to a private non voucher

school, even though these schools are not a realistic option for most students as they serve

the most elite families in Chile. Still, we find very large estimates of the effect of attending

a Matte school. The effect ranges from approximately 14 to 16 points in the language test

                                                                                              22
(0.27-0.32 standard deviations in test scores), and from 25 to 28 points in the math test

(0.47-0.52 standard deviations).

These results taken jointly indicate that SIP schools are able to provide high quality

education to low income students; children at SIP outperform similar students attending

any other type of school. Although the point estimates vary from method to method, they

all suggest that the effect is large and economically and statistically relevant. The effect is

large even when compared to students that belong to the richest families in Chile and that

are enrolled in schools that charge tuition fees that are many times the voucher provided by

the government.

Summing up, given their financial resources, SIP’s methods seem a cost-effective way of

improving the quality of education among low income students. In other words, students

from disadvantaged family backgrounds can achieve high performance. Interestingly, a

similar result has been reported for countries with high PISA results. According to Barber

and Mourshed (2007), there is little correlation between student outcomes and family

background variables in countries with high PISA scores. That is, in the best educational

systems around the world, schools are able to compensate for students’ background

disadvantages.

In what follows, we look further into understanding the estimated differences in

performance. In order to do so, we conducted a series of interviews with principals at SIP

and at other similar schools. In the next section we describe our methodology and results.

                                                                                              23
IV.     A qualitative comparison of SIP and similar schools in Santiago

So far we have shown that students attending SIP schools perform much better than similar

students in other schools, even after controlling for observable characteristics and after

dealing with selection on measured variables. The aim of this section is to identify school

and classroom processes that might explain the reasons that underlie the successful

performance of children at SIP schools. As it was already mentioned, previous research on

factors affecting the effectiveness of school attending low-income students have found that

variables correlated with high achievement are not easy to measure and usually not

observed by an econometrician; thus, a qualitative analysis is needed.16

Therefore, in order to gain some insight on the reasons behind the relative success of

students at SIP schools, we performed a number of interviews with the director of the

Pedagogical Department and with the principals of each school of the network. As a

complement to our statistical analysis based on matching similar students across different

types of schools, we also conducted interviews with principals of schools that serve

populations with similar observable characteristics within the same neighborhoods. We

were able to interview 15 schools, one “neighbor” for each SIP school in Santiago

attending primary level children. The guideline we prepared for the interviews is based on

previous research about effective schools. The questionnaire included questions on the use

of specific teaching methodologies, the goals of the directors, the tasks they engage in, the

16
  See, for instance, Belleï, Muñoz, Pérez and Raczynski (2004); Raczynski and Muñoz (2005), Eyzaguirre
and Fontaine (2008), Pérez and Socias (2008).

                                                                                                         24
characteristics of teachers and students, and school practices in general terms. The

guideline is presented in the appendix.

To choose the schools to interview, we generated a “similarity index”, one for each school

in the neighborhood served by a SIP school. This index was constructed on the basis of

observable characteristics: the vulnerability index of the population at the school,

constructed by the JUNAEB, which summarizes socioeconomic characteristics of the

student’s families; the average family income at the school; maternal years of schooling;

enrollment, and the monthly fee charged by the school.17

Let xijk represent variable i of the vector of characteristics in the index, where j represents

the school, and k the municipality where the school is located. Let xisk be the mean of

variable xi in SIP schools located at municipality k. Finally, let dijk be the difference

between xijk and xisk. Our similarity index is defined as

                                                    n
                                                               1
                                            Ij =∑
                                                   i =1   d ijk − d ik
                                                             σ dik

That is, the index is the sum over i of the inverse of the absolute value of normalized dijk. In

this way, we obtain an index for each school that reaches its highest value for schools that

display characteristics that are most similar to those of the average SIP school in the area.

17
  The JUNAEB, National School Support and Scholarship Board, calculates this vulnerability index at the
school level to target school meal subsidies. The index is based on students’ socioeconomic characteristics.

                                                                                                               25
Once we constructed the index, for each municipality where there are SIP schools we

contacted the principal of the school in the municipality that had the highest index. If we

were unable to contact the principal or to schedule an interview, we contacted the principal

of the school placed second in the similarity ranking, and so on until we were able to

conduct the interview. Table 5 presents the characteristics of the schools of which

principals were interviewed, along with their position in our similarity index.

a. Interviews with Sociedad de Instrucción Primaria

From the analysis of the fifteen SIP primary school interviews it is possible to sketch some

features which help us better understand their academic results.

1.     Principals

Most principals began their careers as classroom teachers, then as department heads, as

deputy principals and finally as principals. The majority has held their positions for a

considerable number of years --the shortest being 6 years, while the majority has over 20

years of experience. Many of the principals have worked in SIP schools and have gained

their administrative training as principals through courses provided by the Sociedad, and by

following postgraduate diplomas and masters programs related to the management of

educational institutions.

The persons below the principals are designated by SIP, mainly by competition, only in few

cases in consultation with the principal. In terms of teachers, most principals regard

themselves as autonomous to hire and fire teachers, although some note that these actions

must be consistent with SIP criteria.

                                                                                              26
In terms of financial management, most principals note that their involvement in financial

decisions has increased recently, although not all claim autonomy in spending decisions.

The principals call attention to the following job characteristics: mission and standards are

clearly defined; they work in teams with SIP’s Pedagogical Department; there is great

commitment, and work is planned jointly and guided by results. Neither principals nor

teachers are immovable from their jobs.

2.     Teachers

Teachers are constantly evaluated on the basis of a number of instruments. One is a self-

evaluation process developed by SIP. Another is the so called ‘formative evaluations’

through which teachers monthly discuss their activities with the principal and the deputy

principal. Finally, class observation is also used; i.e., directors and under directors attend

some lectures in order to provide feedback and help. Moreover, the best evaluated teachers

participate in these observations in order to share the best practices within the school and

even across the network of schools. The results of these evaluations provide feedback to

teachers and the administration, and are also used for the definition of bonuses, training,

and dismissal.

Teachers’ pay consists of a basic salary plus an additional amount for meeting goals. In

addition they receive payments from the general teachers’ incentive scheme established by

the Ministry of Education.

All teachers work for the SIP only and show low rotation and very low absenteeism rates.

The exception is one school where the absenteeism rate is very high and where teachers are

                                                                                                 27
reluctant to take training courses in those periods when there are no classes. Teachers also

undertake additional training and courses both inside and outside SIP.

3.     Students

The pupils begin their schooling at pre-school level and continue there; most of those who

leave the school do so because of a change of residence. In the 6th grade, however, many of

the students hope to enter elite public schools, point at which they also leave the network. A

small fraction of older students leave because they do not adapt to the required level of

work and discipline. Some of the principals interviewed admit that their schools do expel

students on the basis of behavioral or academic problems.

All students have to take a maturity test for kindergarten admission. Moreover, ten out of

the fifteen schools state they have excess demand and thus select their pupils. This selection

process requires students to take a subject knowledge test.

Class attendance is good even if in some cases some students tend to arrive late to school.

Parents and guardians also demonstrate a high attendance record -- over 85 percent-- at

parent meetings and at meetings with specific teachers. However, in two of the schools the

directors report that the school is set up within a highly vulnerable environment and that a

number of parents or guardians are drug addicts or linked to drug traffickers.

Student behavior is recorded in a book that marks positive and negative observations.

Although principals state that they want to reinforce the positive observations – because

they are concerned with student self esteem-, most of the comments are negative. These

annotations are used to inform parents and to supply personality reports. The agreements

                                                                                             28
and commitments between the students, their parents and the school are documented in the

same book.

4.     Methodology used

SIP schools can rely on monthly counseling by the central Pedagogical Department which

is principally delivered as methodological instructions. The schools plan their objectives,

content and activities, and they progress according to the fulfillment of these objectives.

Goals are common across schools and are clearly established. For example, in language

learning, students are expected to read a book per month, learn a given number of new

words per week and spend at least an hour a week in the library. Discipline is also

considered as a key to teaching.

Students normally work in groups using blueprints. The educational process involves

parents, who are periodically informed about the students’ performance, and requested to

participate in parents’ meetings. Workshops are held for parents including teaching them to

read if needed.

A Diagnosis Center – Centro de Diagnóstico Arturo Matte—provides lectures to students

that lag behind at each of the schools in the network. These are reinforcement lectures in

addition to the regular class schedule.

5.     Data driven decision making in schools

SIP schools undertake continuous and systematic evaluations of their students beginning

with a diagnostic test at the start of every school year and then again at the end of the year.

This test, designed by SIP’s Pedagogical Department, provides the necessary information to

                                                                                              29
assess the strengths and weaknesses of students’ learning. The information is shared with

teachers who can change the focus and content of their lectures according to the specific

needs of the students. Subject advisors --internal or external teachers appointed by the

Pedagogical Department-- provide extra support to teachers and classes that need

reinforcement according to the evaluations performed. The tests results are also used to

identify the areas which students that lag behind need to strengthen, and for teachers to talk

to parents and ask them for support and extra involvement. The information is further used

to identify the more advanced students who can benefit from special workshops.

In summary, schools’ academic actions are based on the information gathered from these

periodic evaluations.

b. SIP’s “neighbors”

1.     Principals

There is wide heterogeneity in terms of how long the school director has been in place and

the way the post was filled. In most cases, the owner or the municipality directly appointed

the director; in only three out of fifteen cases the principal reports having competed for the

post. In many private voucher schools, the director is the owner herself or a relative. In

about half the cases, the director knew the school before taking charge, most commonly

working as a teacher.

All interviewed directors have teaching experience -- although not all of them have a

university degree-- and a long trajectory in the educational system, many times in charge of

the Academic Unit (Unidad Técnico Pedagógica or UTP). Two thirds report having

                                                                                             30
advanced studies or formal training in management. Some of them report advanced studies

in the management of educational institutions.

In most cases, the directors do not have the power to choose the directive team that works

with him or her. Contrary to principals at SIP, they cannot appoint or dismiss teachers. It’s

either the municipality or the owner of the school who takes the decisions.

Most principals report having control over the definition of teaching methodologies. Only

four state that this is a process done jointly within the school or discussed with external

authorities, such as a Fundación or the municipality.

When they have control over financial resources, it happens in a very limited manner. They

mention as examples the resources from renting a snack bar or classrooms, which most

likely represent a very small share of all resources available at the school level. Whenever

they refer to the use of resources, they address school maintenance. That is, as opposed to

SIP schools, directors at these schools do not tend to have financial autonomy.

We also asked school directors to describe their typical working day. All start the day

receiving teachers and students and checking that all classes develop normally. Some also

state that they check whether the school is clean. Some follow up over the break or at

students’ lunch time. They all state they spend a portion of the day talking to parents, in

particular those of children that have “critical behavior” and to those of children that arrive

late in a systematic manner. In many times, they state that discipline is the principal’s

responsibility. All spend time at meetings -- held mostly outside the school--, and filling

out reports and surveys mainly sent out by the Ministry. Some teach. All seem to feel that

everyday overall supervision depends upon the director. But some do not seem to like this

                                                                                              31
task. One director states “This is the part of the job I like the least. I used to have a very

dynamic job: creating material, working with teachers,… but now the job is mainly about

meeting people to make sure things actually work”. It is interesting to note that only one of

the interviewed schools has an under-director, so directors may feel they do not have

someone on whom they can delegate a number of tasks.

All coincide with respect to the importance of the UTP coordinator and of the tasks

assigned to him or her. According to the responses of the directors, this person is in charge

of the supervision, the organization, and evaluation of the pedagogical activities. All UTP

coordinators seem to work closely with teachers, revising their material, tests, and results.

All directors assign a crucial role to this unit. One director states that “the UTP coordinator

is the engine of the school in terms of learning and performance”. In a number of cases,

this person also teaches, and is in charge of talking to the parents of children who are not

performing as expected, in particular those at risk of repeating a grade.

We also inquired about whether they feel their directing team is special, and if so, how is it

different from other teams at other schools. Almost all respond that their school is

characterized by a good organizational climate, a climate you do not find in any school.

Directors seem to feel they have good organizational skills and believe that a good climate

is crucial to getting things done. “Organizational climate is more important than

methodologies; things do not work if there is no commitment,” as one states. They talk

about trust, frankness and loyalty as values worth investing on: “we have no quarrels,

everything is stated clearly and frankly, and nobody seems offended,” says another.

                                                                                                 32
Three school directors also state that they are different because their main goal is to attend

the needs of the students. In all these cases, they say that although the children at their

school are poor and live in aggressive environments, their goal is to make children believe

in themselves, and to give them the skills and tools they need to overcome poverty. Two

directors state that an additional goal is to make children feel protected and taken care of

within the school.

Finally, one director states that their school is different because its main goal is to teach

values and a way of life -- fraternity, respect and laboriousness. Another director states that

they are different because they care about extracurricular activities, and that they would like

the children they serve to excel in sports, music and other areas.

These answers strike as quite different to SIP’s stated objectives and method, such as

learning, a common methodology, discipline, and evaluation.

2.     Teachers

Almost all directors state that teachers’ absenteeism rates are extremely low. Teachers do

not go to work only when one of their children or themselves is sick. One director points

out at this fact as a signal of the commitment teachers have with their job. However, in

three cases, the director says that absenteeism rates are high –even up to 20%. Similar to

their counterparts, they interpret these high rates as teachers who do not care much about

the educational project.

In almost all interviewed schools, teachers receive a fixed income, according to the

minimum payment scheme all Chilean teachers in the public sector must receive. This is

                                                                                                33
You can also read